Replay: The EASIEST way to create AI Cover Songs!

welcome back to the channel where we

explore the creative uses of AI and we

have been on a voice cloning and music

kick lately haven't we boys and girls

well I am not going to stop now because

somebody told me about replay now if you

don't know about replay and you have an

interest in voice conversion

specifically taking the voice of any

song and changing it with whatever voice

you'd like to put in there I've been

showing all sorts of tools the past

several months really about how to use

different pieces of software to get

various parts of that task done use this

software to download and separate the

tracks use this software to convert the

audio on your computer and then more

software to put it together and while

this is not a 100% total voice cloning

and conversion solution it goes a long

way to making the process super super

easy and I'm a person who likes to use

tools local on my machine but this is a

tool that I have become completely

addicted to and I spent way too many

hours converting songs all weekend long

I'm going to show you why and I'll bet

you you do it too if you haven't

explored it yet the first thing you need

to do is go get replay and it is free

boys and girls every aspect this project

is free isn't that nice there's no

subscriptions of any kind to do what

we're about to do here it's pretty

mind-blowing actually when you get here

you've got choices you can download it

defaults to Windows on my platform but

you can also download to Mac and Linux

pretty simple install be aware that as

you're running this program it may

occasionally download models from the

internet and slow the process but all

those model downloads are a one-time

operation and it appears that most

everything is taking place on your

machine which is fantastic if I'm wrong

somebody correct me because this is all

about replacing vocals in an audio track

and because I don't want to deal with

copyright restrictions even though I'm

so tempted to play you all the stuff I

did this weekend I just am not going to

take that chance and luckily with sunno

we don't need to worry about that we can

go generate our own song and it's going

to be great so here in sunno I've

already created this song just to

eliminate all the trial and error that's

going to be perfect for our example I'll

play a little bit of it here basically

it was a kuner male solo Jazzy upbeat

swing quartet brass Summer Breeze in the

city I was taking a bunch of Franks and

atra titles and Mish mashing them and

see what they come up with and this is

what we've

[Music]

got okay we won't listen to the whole

thing now so we're going to go ahead and

download the audio track from that

particular song and then we're going to

go to replay now your screen won't look

exactly like this because I've been

filling it up with stuff all weekend

including models that I've downloaded

and I'm going to show you how to do

exactly that but let me just give you a

quick demo of how this basically works

you select any song you'd like music

track with audio on it and drop it right

here or you can record your own vocal

track or the most fun part is entering

in a URL for YouTube and it will

download the video separate the audio do

the conversion for you all in one step

it's really quite amazing as tempted as

I am to play you demos of all the stuff

I did this weekend I just don't want to

risk it so let's just play with what we

did so we're going to drop the song we

just downloaded from sunno right in here

to select or Drop Audio here we can

preview it to make sure it's the right

one it's the right one and now we choose

our our models but Bob I don't have any

modelss I just downloaded this thing

where do I get these models from of

which you speak look at this it says

right here if you just would relax it

says right here looking for more models

yeah 20,000 plus available on weights.

let's click that link right there and

then boom so when you get to weights. G

the first thing you want to do is to

create an account and it's free so all

you got to do is Click login up here and

then you can continue with any of these

ways to get in I generally use Google

once you're in now you can search for

and download these models the easiest

way to do that because the organization

around here is questionable at best just

click on the magnifying glass to search

and just type in the name of the voice

you'd like now they have a lot of

singers here they also have a lot of

characters for example all the SpongeBob

gang is here if I just type sponge

you'll see that and let's say a singer

like Billy Alish if we just type Billy

you'll see that we've got lots of

choices to choose from so you can spend

a lot of time here just finding and

playing with models you can preview what

the model sounds like by the way by just

clicking this Arrow here as a speaking

sample for awaits GGA and it gives you a

speaking sample as you can see we can

dive deep into this side let's get a

voice that I have not used just for the

purposes of this let's get Dean Martin

because he's a kuner and this is a kuner

song let's see if we got Dean Martin up

here we seem to have one let's listen to

it as a speaking sample for awaits GGI

voice model sure sounds enough like him

let's give it a try so we're going to

download the model and what we're going

to get is a ZIP file and when you open

it most of the time you're going to have

something like this sometimes you're

going to have a little bit of a

directory structure but you're going to

have a pth file which is the actual

voice model and you're going to have an

index file which kind of tells the voice

model how to behave first we're going to

extract this into a folder with the rest

of the voice models that we have I'm

actually going to create a folder for

this flect folder and extract once

that's extracted we're going to want to

rename these files because they all come

down as model and that's going to get

quite confusing so let's just rename

this to Dean Martin pth and Dean Martin

index and these are all being saved in

the default model file for replay which

is defined under the app drop down click

on show settings and here's where you

define the current APP directory once we

have that model renamed it's ready to

drag right into replay so we just take

the pth file and literally drag and drop

it right where it says and it says

successfully added Dean Martin and now

Dean Martin shows up in this growing

list of model files that I have and

let's pop in on the settings real quick

because we want to make sure of a few

things stem only means it's going to

skip the voice conversion and output the

vocals only so it's just going to take

the vocals off of this track and create

a file that you can download of the

existing vocals not anything converted

we're not going to do that pre- stemmed

means that the vocals have already been

separated from the audio so if I was to

record my voice here and the record your

own option that was up here I would use

the pre- stemmed option because there's

no music to separate the relative pitch

you change when you're drastically

changing the voices from one to the

other for example if I'm going to change

to a female voice and the female voice

is probably an octave higher than mine I

need to tell Replay that my voice is an

octave lower the easiest way to do that

is just to click on minus1 12 in this

case it's a male voice to a male voice

so we're just going to leave it right

where it is the instrument pitch this

allows you to change the pitch of the

actual music track so if you change the

vocal track to sound a little bit more

real you might need to change the

instrument track to transpose the audio

to meet that vocal you can leave the

stem method here alone and you can leave

all this alone you can take a peek at

advanced settings but I literally have

not changed any of these things except

to set the microphone input value for

when I do record my own thing everything

else I leave just where it is you won't

have Cuda here unless you have an Nvidia

card in here but uh you just go with

whatever you've got you'll probably have

the option of CPU or some sort of GPU if

you have one all right now we're just

going to click on create song now the

first time it separates the audio it

takes a couple of minutes but once

that's done that separated audio is on

your system so it's really easy just to

go through a bunch of other voices and

audition them which we will do so you

can see what's happening here it's

separating the track it's about 16% of

the way through your GPU definitely

determines how fast this goes right now

I'm running this on an Nvidia RTX 2070

super and on another system I run it

with an RTX 390 and it's way faster on

all processes but for the purposes of

this demo I'm using this and it will

take a little bit longer for the track

separation to occur you can by the way

cue jobs while this is doing this I can

go ahead and start another job

completely beautiful dreamer I'm going

to back to replay which we're still

separating the track instead of changing

Bing's voice out with Dean Martin why

don't we change it out with Squidward

all this stays the same and we're going

to go ahead and click on create song and

now you'll see two songs cued now we're

waiting for the first track to finish

separating and do the conversion and

then the other one we'll start while

we're having fun with the first one our

track is finished so if we click on it

up here we'll see that in addition to

the finished conversion we also have the

ability to download The Source tracks we

have the original song track the

converted vocals only the original

vocals only and then the instrumentals

only it's great if we want to remix this

which is what we're going to do let's

just take a listen real quick again to

what the original song sounded

[Music]

like now let listened to it with Dean

Martin's voice put

ink awesome so let's just click remix

again and let's just quickly choose

another voice this time we'll choose a

billy eish voice and because her voice

is higher we're going to go down here to

relative pitch and we're going to click

on plus 12 and we will click on create

song and you'll notice there's no

separation going on we're just loading

in the voice conversion model real quick

changing the voice creating audio files

and it's done a lot quicker already done

to New York in the

summer that's a little high for her

voice what we could do is try and change

the relative pitch and try the

instrumental pitch pitch which I've

never actually done so this is a good

opportunity to do this since that seemed

a little high for her I think the zero

would still be too low let's go down

about six steps here all right but that

means we're going to also have to change

the instrument steps down to six so

let's just see what happens when we do

that definitely

transposed much better

right I think that's great it sounds

pretty good but if we want to sweeten it

up a little bit this is where the beauty

of being able to download those

individual tracks is so now we just go

down here and we just want the

instrumental track and the converted

vocals now that we've got those two

audio tracks I'm just going to download

each of them into my favorite multirack

editor you can use whichever one you

want to provided you can do effects and

other basic editing I have already

created a very simple effects rack with

Reverb that I put the vocal track on and

I've left the music track alone let's

just play it how it is and see how the

levels are hello again this is the

original instrument track now with the

Billy eish sample on

top I took off to New York in the Summer

Breeze I did it my

way that's great let's do the same thing

but let's download the Dean Martin

tracks that we did originally now with

the Dean Martin one we're going to have

to redownload the instrumental track

because that one was not transposed like

we did for the Billy ish ones click that

and now click the converted vocals those

are both being downloaded and I'll just

drag those in like I did before and

click

play I took off to New York in the

summer

RS we don't want to forget about

Squidward and Beautiful Dreamer so let's

just click here we'll download the

instrumental track we'll download the

converted vocals go back into audition

we'll take these out we'll drag the

instrumental in here we'll drag the

vocal in here sometimes when you get

these downloads you'll see that there's

some noise here in the vocal track which

is basically not really singing what

I'll do in my case is I'll just go into

these files make sure that it's noise

and I'm just going to silence this in my

case I'm just going to go under edit

insert silence and click okay and it

just replaces that whole piece with

silence I'm going to do the same thing

here and then let's peek here what this

is yeah that's noise too we also silence

that okay and here in audition those

changes are automatically applied here

in the multitrack so now you can see

that's all cleaned up so let's just go

back to the beginning and hear beautiful

dreamer with Squidward

Tentacles all under fair

use because we are significantly

changing the original by changing the

main vocal here he is now Squidward

Tentacles beautiful

dreamer why for

me Starlight and doom gaps are waiting

for

the I'm your hero right now aren't I so

hopefully your mind is already

sufficiently blown because you could

just sit here and do this all day with

the ease that it is to download these

models and play these out and once you

separate it it's really super fast it's

really cool so have fun with all of that

let's take a look at another little

feature here multimodel what does that

mean select multiple models to use for

one song this is just sort of like a

batch processing let's go back to Summer

Breeze in the city for example and click

remix again click on multimodel I want

Darth Vader and I want farth Brooks

versions of beautiful dreamer because

these are male I need to make sure this

relative pitch is down to zero again

would not make sense to do a mobile

batch with female and a male voice if

the ranges are drastically different and

then I click create song and again all

it's going to do now is it's going to

just convert each of those voices now

what we end up with when we look over

here now we have the number two here now

we have have converted tracks for Frank

Sinatra and G Brooks now let's try

merging models first you have to have

multimodel chosen then we have to choose

the models that we want let's just say

Lily eish and Sheldon Plankton now I'm

going to click on merge and I'm going to

click on create song and see what we get

it's creating an entirely new model from

those weights which didn't take long and

now we have the Summer Breeze in the

city Billy Alis Sheldon Plankton making

sure it's still just one track let's

listen I took off to New York in the

Summer

Breeze I mean how are you going to say

that that's wrong cuz we don't know what

Billy Alish and Squidward would sound

like together this is a 50/50 mix as it

says right here let's see if we've got

the ability to shift who gets how much

yes if I click this icon here I can say

I want more Billy than Sheldon and it

just automatically changes so let's try

it with a preponderance of Billy and

then we'll try it with a preponderance

of Sheldon I took off to New York in the

Summer

Breeze and I did it

no one to so that's odd because I did

mix a female voice and a male voice and

blankon is way down here and she's

somewhere up here maybe if I had done to

12 plus it's okay let's go ahead and

change those ratios and see what we get

bring Billy down and Sheldon Plankton up

took

York that's mostly Plankton maybe I need

to back it up just a little bit more get

a little more Billy in there I did it my

way no one to PE I freaking love this so

much fun now you don't have to just

convert music if you don't want to if

you just want to convert speech it works

perfectly well like that too for example

let's convert my voice

into Squidward

denticles why is my voice constantly

being used for Folly why can't I be

taken seriously as an artist want to

make sure you click that little save

disc there it's not

intuitive why is my voice constantly

being being used for all right that's me

let's go down here to Squidward and this

is where we would click pre stemmed and

when you record your own it

automatically assumes that it's pre-

stemmed it already defaults to zero here

so let's just click on creates on this

should take seconds

yep why is my voice constantly being

used for Folly why can't I be taken

seriously as I'm okay do Darth Vader

I've been going over your records it

seems you are a little late on your TPS

report

I'm hoping that perhaps you can get that

done for me and yeah I'm also going to

need you to come in on

Saturday that' be great I've been going

over your records it seems you are a

little late on your TPS report and yeah

I'm also going to need you to come in on

Saturday so provided you download or can

make your own models this is an amazing

way to quickly change out any voice for

any voice in any song pretty much hands

off except for whatever editing you do

on the other side but this again takes

what you can do with sunno to new levels

because now for any song you create you

can use any singer you want and even

create Duets there's one other feature

that this program has that I'm afraid is

kind of obsolete especially with things

like sunno but this allows you to create

music from a text prompt not songs not

lyrics short little Snippets of audio

probably not more than 10 seconds unless

you have all kinds of time on your hands

and it's not going to be nearly the

quality that you're going to get with

something like sunno at least not yet

these are using models created by meta

slfb to create music from text and

several months ago it was pretty

freaking incredible and there is a lot

you can do with it here since I don't

really think anyone's going to use this

I'll just show you an example of what

I'm talking about right here so let's

say a marching

band pop

ballad with accordians you're going to

click on settings and see that the song

duration default is 10 seconds and if

you haven't downloaded models or if you

have a slower computer this 10-second

clip is going to take you a hot minute

to get and I just don't know that a lot

of people are going to have the patience

for it there are different models that

are being used and downloaded depending

on what you're asking it to do how long

you want it to be whether or not you're

trying to guide it with a Melody every

time you activate a new model it

downloads it takes a while but it's a

onetime deal and then it all happens on

your system right now let's just say a

marching band pop ballad with accordion

create music running onetime setup which

means I've asked it to download a model

which I don't get have on my system so

that's going to take a little bit of

time you'll notice that once it actually

starts doing the conversion we get a

countdown here 45 out of 500 steps and

truly these days the payoff just isn't

worth sitting here and doing this but I

want to show you anyway cuz it's in here

let's see what we

[Applause]

got now let's just for fun go into sunno

and give it the exact same prompt I'm

going to go into custom mode I'm going

to say instrumental and the the style is

exactly what I did there a marching band

pop ballad with accordians and click

create no freaking contest I am not

kidding when I say I have spent hours

with replay this weekend just

downloading things from YouTube changing

out the voices laughing my ass off and

doing it again and the quality of the

separations is really really good if you

isolate the vocal tracks They sound

amazing so have a ton of fun with this

I'd love to hear what you do with it if

you enjoy this type of material anything

Ai and creativity related and going down

rabbit holes and Mish mashing tools

together and being a real mad scientist

about it I invite you to subscribe if

you subscribe now I will not look for

you I will not pursue you but if you do

not I will look for you I will find you

and

I

[Music]